DAS: Densely-Anchored Sampling for Deep Metric Learning

Liu, Lizhao; Huang, Shangxin; Zhuang, Zhuangwei; Tan, Mingkui; Wang, Yaowei

doi:10.1007/978-3-031-19809-0_23

Cited by 12 publications

(3 citation statements)

References 54 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Influenced by XBM, some work has embedded the memory bank into the network to adapt various tasks [40], [86]- [107]. To enhance the diversity of semantic differences, Liu et al [99] construct a memory bank using historical intra-class embedding representations. Motivated by the issue of slow feature drift, Wang et al [100] propose the quantization code memory bank to lower feature drift to use historical feature representation effectively.…”

Section: Aspect Of Historical Typementioning

confidence: 99%

A Survey of Historical Learning: Learning Models with Learning History

Li¹,

Ge²,

Yang³

et al. 2023

Preprint

View full text Add to dashboard Cite

New knowledge originates from the old. The various types of elements, deposited in the training history, are a large amount of wealth for improving learning deep models. In this survey, we comprehensively review and summarize the topic-"Historical Learning: Learning Models with Learning History", which learns better neural models with the help of their learning history during its optimization, from three detailed aspects: Historical Type (what), Functional Part (where) and Storage Form (how). To our best knowledge, it is the first survey that systematically studies the methodologies which make use of various historical statistics when training deep neural networks. The discussions with related topics like recurrent/memory networks, ensemble learning, and reinforcement learning are demonstrated. We also expose future challenges of this topic and encourage the community to pay attention to the think of historical learning principles when designing algorithms. The paper list related to historical learning is available at https://github.com/Martinser/Awesome-Historical-Learning.

show abstract

Section: Aspect Of Historical Typementioning

confidence: 99%

A Survey of Historical Learning: Learning Models with Learning History

Li¹,

Ge²,

Yang³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…These methods use linear interpolation to generate new samples, leading to false negatives and positives due to the difficulties in determining class boundaries on the manifold structure. DAS [21] and IAA [5] try to solve this issue by adjusting features slightly and adding extra examples during midtraining of the model, aiming to make sure that new examples still belong to the same class. Despite these attempts, completely removing this problem remains challenging.…”

Section: Hard Sample Miningmentioning

confidence: 99%

“…Approaches like hard sample mining and generation have been proposed to aid network convergence by introducing a substantial gradient [1,41,12,34,49,48,17,18]. Recently, an advancement in generating appropriate hard samples involves creating supplementary training data [17,38,5,21]. This is primarily * Corresponding to:csong@zjsu.edu.cn achieved through linear interpolation, a prevalent method for generating synthetic samples.…”

Section: Introductionmentioning

confidence: 99%

HSE: Hybrid Species Embedding for Deep Metric Learning

Yang,

Sun,

et al. 2023

2023 IEEE/CVF International Conference on Computer Vision (ICCV)

View full text Add to dashboard Cite

Deep metric learning is crucial for finding an embedding function that can generalize to training and testing data, including unknown test classes. However, limited training samples restrict the model's generalization to downstream tasks. While adding new training samples is a promising solution, determining their labels remains a significant challenge. Here, we introduce Hybrid Species Embedding (HSE), which employs mixed sample data augmentations to generate hybrid species and provide additional training signals. We demonstrate that HSE outperforms multiple stateof-the-art methods in improving the metric Recall@K on the CUB-200 , CAR-196 and SOP datasets, thus offering a novel solution to deep metric learning's limitations.

show abstract